Sallen-Key filter problem

[Nico Coesel]

Hello all,

I'm having a weird problem with a Sallen-Key filter. I've build a
filter which should have a cut-off frequency around 50MHz. I'm using a
current-feedback opamp which has a 1GHz bandwidth. The problem seems
to be in the feedback capacitor. The circuits acts like if the
feedback capacitor is twice the actual value so the pole is in the
wrong place.

If I put a 33pf capacitor in the real life circuit a simulation which
uses a 66pf feedback capacitor gives me the same results as the real
life circuit. I also tried a 68pf capacitor.

The filter is driven by a unity gain buffer. I'm using a 1:20 probe
with a 50 Ohm output (passive divider) on an oscilloscope with a 50
Ohm input. The signal going into the filter looks fine. The whole
circuit is on a 4 layer PCB.

Does anyone have some thoughts?

 

[Helmut Sennewald]

Hello all,

I'm having a weird problem with a Sallen-Key filter. I've build a
filter which should have a cut-off frequency around 50MHz. I'm using a
current-feedback opamp which has a 1GHz bandwidth. The problem seems
to be in the feedback capacitor. The circuits acts like if the
feedback capacitor is twice the actual value so the pole is in the
wrong place.

If I put a 33pf capacitor in the real life circuit a simulation which
uses a 66pf feedback capacitor gives me the same results as the real
life circuit. I also tried a 68pf capacitor.

The filter is driven by a unity gain buffer. I'm using a 1:20 probe
with a 50 Ohm output (passive divider) on an oscilloscope with a 50
Ohm input. The signal going into the filter looks fine. The whole
circuit is on a 4 layer PCB.

Does anyone have some thoughts?

Hello Nico,

Have you added some capacitors for the PCB-traces into the schematic?

Can you show your SPICE netlist or even better the schematic (.asc if
LTspice)?
If you don't want post it, you can send it to my email address.
What's your simulator?

Best regards,
Helmut

 

[MooseFET]

Hello Nico,

Have you added some capacitors for the PCB-traces into the schematic?

Can you show your SPICE netlist or even better the schematic (.asc if
LTspice)?
If you don't want post it, you can send it to my email address.
What's your simulator?

Also:

What is the gain and phase of the op-amp at the cut off frequency. A
good rule of thumb is that to get 10% accuracy on the numbers:

GBP = 10 * F * Q^2

F = the cut off frequency
Q = the intended Q
GBP = the gain bandwidth product

GBP needs to be the value at the cut off frequency. Many op-amps have
a GBP that decreases with frequency.

Also also:
Beware of circuit inductances. 100nH is enough to mess up the circuit.

 

[Nico Coesel]

Hello Nico,

Have you added some capacitors for the PCB-traces into the schematic?

No not really. I did change , but the layout is very tight. The opamp
is in an SO8 package the components for the sallen-key filter are as
close to the opamp as possible. I did some more tests. If I change the
feedback capacitor to 1nf and the capacitor to ground into 100pf the
circuit works as simulated.

There is definitely something going on with parasitics here.

I also replaced the opamp just to make sure the opamp isn't defective.

Can you show your SPICE netlist or even better the schematic (.asc if
LTspice)?

It is the basic sallen-key filter like on:
http://en.wikipedia.org/wiki/Sallen_Key_filter
with R1=R2=120 Ohms. C1 is 68pf. C2 is 6.8pf in the real circuit and
9pf in the simulation to compensate for parasitic capacitances. But
the 9pf capacitor only moves the zero. The pole is almost unaffected.

The filter is driven by a MAX4200 unity gain buffer. The filter itself
is build around MAX4225 current feedback opamp. The load is only the
probe (1k parallel with 2pf to ground). Power supply is +/-5V.

 

[Helmut Sennewald]

No not really. I did change , but the layout is very tight. The opamp
is in an SO8 package the components for the sallen-key filter are as
close to the opamp as possible. I did some more tests. If I change the
feedback capacitor to 1nf and the capacitor to ground into 100pf the
circuit works as simulated.

There is definitely something going on with parasitics here.

I also replaced the opamp just to make sure the opamp isn't defective.


It is the basic sallen-key filter like on:
http://en.wikipedia.org/wiki/Sallen_Key_filter
with R1=R2=120 Ohms. C1 is 68pf. C2 is 6.8pf in the real circuit and
9pf in the simulation to compensate for parasitic capacitances. But
the 9pf capacitor only moves the zero. The pole is almost unaffected.

Hello Nico,

A Butterworth filter response will require C1=2*C2

Why do you have such a large ratio C1/C2?
This will result in a lot of gain peaking near the corner frequency.

I have simulated it with LTspice. This filter requires C1=36p and C2=18p
for a 50MHz bandwidth with a maximum flat response(Butterworth).

The most critical node is the minus input of the MAX4225 because
off the large 560hm feedback resistor in this path. Even 1pF will
shift the corner frequency by 12MHz towards higher frequencies.
This shift can be compensated with a much lower feedback
capacitor C1. That's exactly what you have done in your real circuit.

Best regards,
Helmut

 

[Jim Thompson]

[snip]

The filter is driven by a MAX4200 unity gain buffer. The filter itself
is build around MAX4225 current feedback opamp. The load is only the
probe (1k parallel with 2pf to ground). Power supply is +/-5V.

Schematic? What makes you think that a Sallen-Key can be just dropped
into a current-feedback device?

Sallen-Key sucketh the big lemon ;-)

...Jim Thompson

 

[[email protected]]

[snip]

The filter is driven by a MAX4200 unity gain buffer. The filter itself
is build around MAX4225 current feedback opamp. The load is only the
probe (1k parallel with 2pf to ground). Power supply is +/-5V.

Schematic? What makes you think that a Sallen-Key can be just dropped
into a current-feedback device?

Sallen-Key sucketh the big lemon ;-)

...Jim Thompson
--
| James E.Thompson, P.E. | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona Voice480)460-2350 | |
| E-mail Address at Website Fax480)460-2142 | Brass Rat |
| http://www.analog-innovations.com | 1962 |

America: Land of the Free, Because of the Brave

Generally you try to limit your high impedance nodes if you want to be
immune to stray capacitance. This is something the Salen and Key
doesn't do. If this were audio frequency design, I'd say pick a
topology that uses one op amp per pole. These generally reduce the
high impedance nodes. However, at 50MHz, I just don't know the best
solution. More op amps means more phase shift.

The integrated video filter designs tend to use gm-c topology, but
that might be because it integrates well.

 

[MooseFET]

Hello Nico,

A Butterworth filter response will require C1=2*C2

Why do you have such a large ratio C1/C2?
This will result in a lot of gain peaking near the corner frequency.

It also makes the circuit sensitive to the phase shift in the op-amp.
I looked at the data sheet and the -3dB point is about 150MHz. This
usually means that the phase will start running late at 15MHz.

 

[John Larkin]

[snip]

The filter is driven by a MAX4200 unity gain buffer. The filter itself
is build around MAX4225 current feedback opamp. The load is only the
probe (1k parallel with 2pf to ground). Power supply is +/-5V.

Schematic? What makes you think that a Sallen-Key can be just dropped
into a current-feedback device?

A cfb opamp makes a perfectly fine follower, which is all a s-k filter
needs.

Sallen-Key sucketh the big lemon ;-)

They are nice and simple, and have tolerable sensitivity as long as
you're not after high-q, many-stage filters. They are fine for things
like 2-4 pole Gaussians or 2-pole Butterworths.

You can get a 3-pole lowpass from one opamp with a bit of fudging.

I don't think the op has a problem... just use the cap that works!

John

 

[Helmut Sennewald]

It also makes the circuit sensitive to the phase shift in the op-amp.
I looked at the data sheet and the -3dB point is about 150MHz. This
usually means that the phase will start running late at 15MHz.

Hello Ken,

The MAX4225 has the -3dB corner at 1GHZ for a gain of +1..
I wonder how you read the datashet.

Best regards,
Helmut

 

[Genome]

Hello all,

I'm having a weird problem with a Sallen-Key filter. I've build a
filter which should have a cut-off frequency around 50MHz. I'm using a
current-feedback opamp which has a 1GHz bandwidth. The problem seems
to be in the feedback capacitor. The circuits acts like if the
feedback capacitor is twice the actual value so the pole is in the
wrong place.

If I put a 33pf capacitor in the real life circuit a simulation which
uses a 66pf feedback capacitor gives me the same results as the real
life circuit. I also tried a 68pf capacitor.

The filter is driven by a unity gain buffer. I'm using a 1:20 probe
with a 50 Ohm output (passive divider) on an oscilloscope with a 50
Ohm input. The signal going into the filter looks fine. The whole
circuit is on a 4 layer PCB.

Does anyone have some thoughts?

A little knowledge is a dangerous thing... however it might make you
wonder.

Yes I do not know what I am blithering about, I'm only guessing.

For an equal component LC section then 50MHz is 3.18nH and 3.18nF and
for some sort of damping...... wierd....... you want a termination
resistance of 1R. What were the chances of that?

Anyway let's say that is boring and make L 100nH and C 100p for a
termination thing of about ten times Pi ohms or 33R.

Your C is in the old blokes range of ceramic 1% NPO stuff so now all
you need is your L.

http://www.emclab.umr.edu/pcbtlc/microstrip.html

I don't have a basic clue.... but if you fiddle about with that you
might get some idea about something.....

I don't know.

How quick would it be to throw up a PCB with multiple test sections of
some sort of length track with 100pF caps at the end of them and see
what happens......?

DNA

 

[Nico Coesel]

Hello all,

I'm having a weird problem with a Sallen-Key filter. I've build a
filter which should have a cut-off frequency around 50MHz. I'm using a
current-feedback opamp which has a 1GHz bandwidth. The problem seems
to be in the feedback capacitor. The circuits acts like if the
feedback capacitor is twice the actual value so the pole is in the
wrong place.

If I put a 33pf capacitor in the real life circuit a simulation which
uses a 66pf feedback capacitor gives me the same results as the real
life circuit. I also tried a 68pf capacitor.

The filter is driven by a unity gain buffer. I'm using a 1:20 probe
with a 50 Ohm output (passive divider) on an oscilloscope with a 50
Ohm input. The signal going into the filter looks fine. The whole
circuit is on a 4 layer PCB.

Does anyone have some thoughts?

I think I've found the problem. Don't start laughing ... a bad solder
joint on the first (0603) resistor added about 80 Ohms. Adding 80 Ohms
or doubling the feedback capacitor has almost the same effect on the
transfer characteristic of my filter setup.

Now the ideal simulation result is just a few MHz off which is
acceptable (to me). Pfew!

Thanks for thinking along!

Nico

 

[Michael A. Terrell]

I think I've found the problem. Don't start laughing ... a bad solder
joint on the first (0603) resistor added about 80 Ohms. Adding 80 Ohms
or doubling the feedback capacitor has almost the same effect on the
transfer characteristic of my filter setup.

Now the ideal simulation result is just a few MHz off which is
acceptable (to me). Pfew!

Thanks for thinking along!

that was a common problem on machines soldered resistors on the
Sallen Key video low pass filters we built. The engineer of that board
chose to use 100 PF for all the capacitors. Quite often, just swapping
the pair of 1% capacitors in the first stage would pull a slightly out
of tolerance filter into spec. Using 14, 1% tolerance parts didn't
always hit the 10% tolerance for each filter. We used half of an LF347
up to 2 MHz, and current op amps above that. The entire video system was
75 ohms, to the output stage, then it could be configured for 50 or 75
Ohm output impedance. I don't have any more details that I can give,
because of the NDA.

--
Service to my country? Been there, Done that, and I've got my DD214 to
prove it.
Member of DAV #85.

Michael A. Terrell
Central Florida

 

[MooseFET]

Hello Ken,

The MAX4225 has the -3dB corner at 1GHZ for a gain of +1..
I wonder how you read the datashet.

Hmmmm I can't figure it out either. I must have been looking at some
other part's data sheet and didn't notice.